Computational Methods for Elastic Wave-Mode Separation
Mode Separation; Elastic Full-Waveform Inversion; Machine Learning; Physics Informed Neural Networks; Helmholtz Decomposition
Mode conversion in non-homogeneous elastic media complicates the accurate interpretation of physical properties. Decomposing these modes correctly is crucial across various areas of Physics. In particular, Elastic Full-Waveform Inversion (EFWI) can achieve improved results when wave modes are accurately separated, since reliable P- and S-wave information can mitigate crosstalk. In this work, we investigate different computational methods for separating P- and S-modes using the Helmholtz decomposition. To address this problem, we first carefully derive an analytical solution for the elastic wave equation in homogeneous media. After this, we revisit standard methods for wave mode separation. Among these separation approaches, the scalar formulation offers a dimensionally scalable reduction in computational cost compared to the traditional vector formulation. Since Machine Learning (ML) has been used to address this problem, we also verify the capabilities of a Physics-Informed Neural Network (PINN) to separate elastic modes in both homogeneous and non-homogeneous complex media by solving a scalar Poisson equation. The results for separated modes using PINN closely match those from conventional numerical techniques while exhibiting reduced transverse-wave leakage. This work contributes to the application of physics-informed machine learning methods to describe aspects of wave phenomena.